-
Notifications
You must be signed in to change notification settings - Fork 747
/
data.go
344 lines (321 loc) · 12.3 KB
/
data.go
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
// Copyright 2020 gorse Project Authors
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
package ranking
import (
"bufio"
"fmt"
"github.com/juju/errors"
"github.com/scylladb/go-set"
"github.com/scylladb/go-set/i32set"
"github.com/scylladb/go-set/strset"
"github.com/zhenghaoz/gorse/base"
"github.com/zhenghaoz/gorse/base/encoding"
"github.com/zhenghaoz/gorse/base/log"
"github.com/zhenghaoz/gorse/model"
"go.uber.org/zap"
"os"
"reflect"
"strings"
)
// DataSet contains preprocessed data structures for recommendation models.
type DataSet struct {
UserIndex base.Index
ItemIndex base.Index
FeedbackUsers base.Array[int32]
FeedbackItems base.Array[int32]
UserFeedback [][]int32
ItemFeedback [][]int32
Negatives [][]int32
ItemLabels [][]int32
UserLabels [][]int32
HiddenItems []bool
ItemCategories [][]string
CategorySet *strset.Set
// statistics
NumItemLabels int32
NumUserLabels int32
NumItemLabelUsed int
NumUserLabelUsed int
}
// NewMapIndexDataset creates a data set.
func NewMapIndexDataset() *DataSet {
s := new(DataSet)
s.CategorySet = strset.New()
// Create index
s.UserIndex = base.NewMapIndex()
s.ItemIndex = base.NewMapIndex()
// Initialize slices
s.UserFeedback = make([][]int32, 0)
s.ItemFeedback = make([][]int32, 0)
return s
}
func NewDirectIndexDataset() *DataSet {
dataset := new(DataSet)
// Create index
dataset.UserIndex = base.NewDirectIndex()
dataset.ItemIndex = base.NewDirectIndex()
// Initialize slices
dataset.UserFeedback = make([][]int32, 0)
dataset.ItemFeedback = make([][]int32, 0)
dataset.Negatives = make([][]int32, 0)
return dataset
}
func (dataset *DataSet) Bytes() int {
var bytes uintptr
bytes += uintptr(dataset.UserIndex.Bytes())
bytes += uintptr(dataset.ItemIndex.Bytes())
bytes += uintptr(dataset.FeedbackUsers.Bytes())
bytes += uintptr(dataset.FeedbackItems.Bytes())
// UserFeedback + ItemFeedback + Negatives
bytes += reflect.TypeOf(dataset.UserFeedback).Elem().Size() * uintptr(len(dataset.UserFeedback)+len(dataset.ItemFeedback))
bytes += reflect.TypeOf(dataset.UserFeedback).Elem().Elem().Size() * uintptr(dataset.Count()*2)
bytes += encoding.MatrixBytes(dataset.Negatives)
// ItemLabels + UserLabels
bytes += reflect.TypeOf(dataset.ItemLabels).Elem().Size() * uintptr(len(dataset.ItemLabels)+len(dataset.UserLabels))
bytes += reflect.TypeOf(dataset.ItemLabels).Elem().Elem().Size() * uintptr(dataset.NumItemLabelUsed+dataset.NumUserLabelUsed)
bytes += encoding.ArrayBytes(dataset.HiddenItems)
bytes += encoding.ArrayBytes(dataset.ItemCategories)
return int(bytes)
}
func (dataset *DataSet) AddUser(userId string) {
dataset.UserIndex.Add(userId)
userIndex := dataset.UserIndex.ToNumber(userId)
for int(userIndex) >= len(dataset.UserFeedback) {
dataset.UserFeedback = append(dataset.UserFeedback, make([]int32, 0))
}
}
func (dataset *DataSet) AddItem(itemId string) {
dataset.ItemIndex.Add(itemId)
itemIndex := dataset.ItemIndex.ToNumber(itemId)
for int(itemIndex) >= len(dataset.ItemFeedback) {
dataset.ItemFeedback = append(dataset.ItemFeedback, make([]int32, 0))
}
}
func (dataset *DataSet) AddFeedback(userId, itemId string, insertUserItem bool) {
if insertUserItem {
dataset.UserIndex.Add(userId)
}
if insertUserItem {
dataset.ItemIndex.Add(itemId)
}
userIndex := dataset.UserIndex.ToNumber(userId)
itemIndex := dataset.ItemIndex.ToNumber(itemId)
if userIndex != base.NotId && itemIndex != base.NotId {
dataset.FeedbackUsers.Append(userIndex)
dataset.FeedbackItems.Append(itemIndex)
for int(itemIndex) >= len(dataset.ItemFeedback) {
dataset.ItemFeedback = append(dataset.ItemFeedback, make([]int32, 0))
}
dataset.ItemFeedback[itemIndex] = append(dataset.ItemFeedback[itemIndex], userIndex)
for int(userIndex) >= len(dataset.UserFeedback) {
dataset.UserFeedback = append(dataset.UserFeedback, make([]int32, 0))
}
dataset.UserFeedback[userIndex] = append(dataset.UserFeedback[userIndex], itemIndex)
}
}
func (dataset *DataSet) SetNegatives(userId string, negatives []string) {
userIndex := dataset.UserIndex.ToNumber(userId)
if userIndex != base.NotId {
for int(userIndex) >= len(dataset.Negatives) {
dataset.Negatives = append(dataset.Negatives, make([]int32, 0))
}
dataset.Negatives[userIndex] = make([]int32, 0, len(negatives))
for _, itemId := range negatives {
itemIndex := dataset.ItemIndex.ToNumber(itemId)
if itemIndex != base.NotId {
dataset.Negatives[userIndex] = append(dataset.Negatives[userIndex], itemIndex)
}
}
}
}
func (dataset *DataSet) Count() int {
if dataset.FeedbackUsers.Len() != dataset.FeedbackItems.Len() {
panic("dataset.FeedbackUsers.Len() != dataset.FeedbackItems.Len()")
}
return dataset.FeedbackUsers.Len()
}
// UserCount returns the number of UserFeedback.
func (dataset *DataSet) UserCount() int {
return int(dataset.UserIndex.Len())
}
// ItemCount returns the number of ItemFeedback.
func (dataset *DataSet) ItemCount() int {
return int(dataset.ItemIndex.Len())
}
func createSliceOfSlice(n int) [][]int32 {
x := make([][]int32, n)
for i := range x {
x[i] = make([]int32, 0)
}
return x
}
func (dataset *DataSet) NegativeSample(excludeSet *DataSet, numCandidates int) [][]int32 {
if len(dataset.Negatives) == 0 {
rng := base.NewRandomGenerator(0)
dataset.Negatives = make([][]int32, dataset.UserCount())
for userIndex := 0; userIndex < dataset.UserCount(); userIndex++ {
s1 := set.NewInt32Set(dataset.UserFeedback[userIndex]...)
s2 := set.NewInt32Set(excludeSet.UserFeedback[userIndex]...)
dataset.Negatives[userIndex] = rng.SampleInt32(0, int32(dataset.ItemCount()), numCandidates, s1, s2)
}
}
return dataset.Negatives
}
// Split dataset by user-leave-one-out method. The argument `numTestUsers` determines the number of users in the test
// set. If numTestUsers is equal or greater than the number of total users or numTestUsers <= 0, all users are presented
// in the test set.
func (dataset *DataSet) Split(numTestUsers int, seed int64) (*DataSet, *DataSet) {
trainSet, testSet := new(DataSet), new(DataSet)
trainSet.NumItemLabels, testSet.NumItemLabels = dataset.NumItemLabels, dataset.NumItemLabels
trainSet.NumUserLabels, testSet.NumUserLabels = dataset.NumUserLabels, dataset.NumUserLabels
trainSet.HiddenItems, testSet.HiddenItems = dataset.HiddenItems, dataset.HiddenItems
trainSet.ItemCategories, testSet.ItemCategories = dataset.ItemCategories, dataset.ItemCategories
trainSet.CategorySet, testSet.CategorySet = dataset.CategorySet, dataset.CategorySet
trainSet.ItemLabels, testSet.ItemLabels = dataset.ItemLabels, dataset.ItemLabels
trainSet.UserLabels, testSet.UserLabels = dataset.UserLabels, dataset.UserLabels
trainSet.NumItemLabelUsed, testSet.NumItemLabelUsed = dataset.NumItemLabelUsed, dataset.NumItemLabelUsed
trainSet.NumUserLabelUsed, testSet.NumUserLabelUsed = dataset.NumUserLabelUsed, dataset.NumUserLabelUsed
trainSet.UserIndex, testSet.UserIndex = dataset.UserIndex, dataset.UserIndex
trainSet.ItemIndex, testSet.ItemIndex = dataset.ItemIndex, dataset.ItemIndex
trainSet.UserFeedback, testSet.UserFeedback = createSliceOfSlice(dataset.UserCount()), createSliceOfSlice(dataset.UserCount())
trainSet.ItemFeedback, testSet.ItemFeedback = createSliceOfSlice(dataset.ItemCount()), createSliceOfSlice(dataset.ItemCount())
rng := base.NewRandomGenerator(seed)
if numTestUsers >= dataset.UserCount() || numTestUsers <= 0 {
for userIndex := int32(0); userIndex < int32(dataset.UserCount()); userIndex++ {
if len(dataset.UserFeedback[userIndex]) > 0 {
k := rng.Intn(len(dataset.UserFeedback[userIndex]))
testSet.FeedbackUsers.Append(userIndex)
testSet.FeedbackItems.Append(dataset.UserFeedback[userIndex][k])
testSet.UserFeedback[userIndex] = append(testSet.UserFeedback[userIndex], dataset.UserFeedback[userIndex][k])
testSet.ItemFeedback[dataset.UserFeedback[userIndex][k]] = append(testSet.ItemFeedback[dataset.UserFeedback[userIndex][k]], userIndex)
for i, itemIndex := range dataset.UserFeedback[userIndex] {
if i != k {
trainSet.FeedbackUsers.Append(userIndex)
trainSet.FeedbackItems.Append(itemIndex)
trainSet.UserFeedback[userIndex] = append(trainSet.UserFeedback[userIndex], itemIndex)
trainSet.ItemFeedback[itemIndex] = append(trainSet.ItemFeedback[itemIndex], userIndex)
}
}
}
}
} else {
testUsers := rng.SampleInt32(0, int32(dataset.UserCount()), numTestUsers)
for _, userIndex := range testUsers {
if len(dataset.UserFeedback[userIndex]) > 0 {
k := rng.Intn(len(dataset.UserFeedback[userIndex]))
testSet.FeedbackUsers.Append(userIndex)
testSet.FeedbackItems.Append(dataset.UserFeedback[userIndex][k])
testSet.UserFeedback[userIndex] = append(testSet.UserFeedback[userIndex], dataset.UserFeedback[userIndex][k])
testSet.ItemFeedback[dataset.UserFeedback[userIndex][k]] = append(testSet.ItemFeedback[dataset.UserFeedback[userIndex][k]], userIndex)
for i, itemIndex := range dataset.UserFeedback[userIndex] {
if i != k {
trainSet.FeedbackUsers.Append(userIndex)
trainSet.FeedbackItems.Append(itemIndex)
trainSet.UserFeedback[userIndex] = append(trainSet.UserFeedback[userIndex], itemIndex)
trainSet.ItemFeedback[itemIndex] = append(trainSet.ItemFeedback[itemIndex], userIndex)
}
}
}
}
testUserSet := i32set.New(testUsers...)
for userIndex := int32(0); userIndex < int32(dataset.UserCount()); userIndex++ {
if !testUserSet.Has(userIndex) {
for _, itemIndex := range dataset.UserFeedback[userIndex] {
trainSet.FeedbackUsers.Append(userIndex)
trainSet.FeedbackItems.Append(itemIndex)
trainSet.UserFeedback[userIndex] = append(trainSet.UserFeedback[userIndex], itemIndex)
trainSet.ItemFeedback[itemIndex] = append(trainSet.ItemFeedback[itemIndex], userIndex)
}
}
}
}
return trainSet, testSet
}
// GetIndex gets the i-th record by <user index, item index, rating>.
func (dataset *DataSet) GetIndex(i int) (int32, int32) {
return dataset.FeedbackUsers.Get(i), dataset.FeedbackItems.Get(i)
}
func loadTest(dataset *DataSet, path string) error {
// Open
file, err := os.Open(path)
if err != nil {
return errors.Trace(err)
}
defer func(file *os.File) {
err = file.Close()
if err != nil {
log.Logger().Error("failed to close file", zap.Error(err))
}
}(file)
// Read lines
scanner := bufio.NewScanner(file)
for scanner.Scan() {
line := scanner.Text()
fields := strings.Split(line, "\t")
positive, negative := fields[0], fields[1:]
if positive[0] != '(' || positive[len(positive)-1] != ')' {
return fmt.Errorf("wrong foramt: %v", line)
}
positive = positive[1 : len(positive)-1]
fields = strings.Split(positive, ",")
userId, itemId := fields[0], fields[1]
dataset.AddFeedback(userId, itemId, true)
dataset.SetNegatives(userId, negative)
}
return scanner.Err()
}
func loadTrain(path string) (*DataSet, error) {
// Open
file, err := os.Open(path)
if err != nil {
return nil, err
}
defer func(file *os.File) {
err = file.Close()
if err != nil {
log.Logger().Error("failed to close file", zap.Error(err))
}
}(file)
// Read lines
scanner := bufio.NewScanner(file)
dataset := NewDirectIndexDataset()
for scanner.Scan() {
line := scanner.Text()
fields := strings.Split(line, "\t")
userId, itemId := fields[0], fields[1]
dataset.AddFeedback(userId, itemId, true)
}
return dataset, scanner.Err()
}
// LoadDataFromBuiltIn loads a built-in Data set. Now support:
func LoadDataFromBuiltIn(dataSetName string) (*DataSet, *DataSet, error) {
// Extract Data set information
trainFilePath, testFilePath, err := model.LocateBuiltInDataset(dataSetName, model.FormatNCF)
if err != nil {
return nil, nil, err
}
// Load dataset
trainSet, err := loadTrain(trainFilePath)
if err != nil {
return nil, nil, err
}
testSet := NewDirectIndexDataset()
testSet.UserIndex = trainSet.UserIndex
testSet.ItemIndex = trainSet.ItemIndex
err = loadTest(testSet, testFilePath)
if err != nil {
return nil, nil, err
}
return trainSet, testSet, nil
}